Powers of Two as Sums of Two Lucas Numbers

1 3 47 6 3 11 Journal of Integer Sequences, Vol. 17 (014), Article 14.8.3 Powers of Two as Sums of Two Lucas Numbers Jhon J. Bravo Mathematics Department University of Cauca Street 5 No. 4-70 Popayán, Cauca Colombia jbravo@unicauca.edu.co Florian Luca School of Mathematics University of the Witwatersrand P. O. Box Wits 050 South Africa and Mathematical Institute UNAM Juriquilla Santiago de Querétaro 7630 Querétaro de Arteaga Mexico fluca@matmor.unam.mx Abstract Let (L n ) n 0 be the Lucas sequence given by L 0 =, L 1 = 1 and L n+ = L n+1 +L n for n 0. In this paper, we are interested in finding all powers of two which are sums of two Lucas numbers, i.e., we study the Diophantine equation L n +L m = a in 1

nonnegative integers n, m, and a. The proof of our main theorem uses lower bounds for linear forms in logarithms, properties of continued fractions, and a version of the Baker- Davenport reduction method in diophantine approximation. This paper continues our previous work where we obtained a similar result for the Fibonacci numbers. 1 Introduction Let (F n ) n 0 be the Fibonacci sequence given by F 0 = 0, F 1 = 1 and F n+ = F n+1 + F n for all n 0. The Fibonacci numbers are famous for possessing wonderful and amazing properties. They are accompanied by the sequence of Lucas numbers, which is as important as the Fibonacci sequence. The Lucas sequence (L n ) n 0 follows the same recursive pattern as the Fibonacci numbers, but with initial conditions L 0 = and L 1 = 1. The study of properties of the terms of such sequences, or more generally, linear recurrence sequences, has a very long history and has generated a huge literature. For the beauty and rich applications of these numbers and their relatives, one can see Koshy s book [9]. For example, a remarkable property of the Fibonacci sequence is that 1, and 8 are the only Fibonacci numbers which are powers of. One proof of this fact follows from Carmichael s primitive divisor theorem [6], which states that for n greater than 1, the nth Fibonacci number F n has at least one prime factor that is not a factor of any previous Fibonacci number (see the paper of Bilu, Hanrot, and Voutier [] for the most general version of the above statement). Similarly, it is well known that 1, and 4 are the only powers of that appear in the Lucas sequence. The problem of finding all perfect powers in the Fibonacci sequence and the Lucas sequencewasafamousopenproblemfinallysolvedin006inapaperinannals of Mathematics by Bugeaud, Mignotte, and Siksek [3]. In their work, they applied a combination of Baker s method, the modular approach and some classical techniques to show that the only perfect powers in the Fibonacci sequence are 0, 1, 8 and 144, and the only perfect powers in the Lucas sequence are 1 and 4. A detailed account of this problem can be found in [3, Section 10]. In our recent paper [5], we found all powers of which are the sums of at most two Fibonacci numbers. Specifically, we proved the following. Theorem 1. The only solutions of the Diophantine equation F n + F m = a in positive integers n,m and a with n m are given by F 1 =, F =, F 3 = 4, F 6 = 16, and F +F 1 =, F 4 +F 1 = 4, F 4 +F = 4, F 5 +F 4 = 8, F 7 +F 4 = 16. Inthispaper,weproveananalogueofTheorem1whenthesequenceofFibonaccinumbers is replaced by the sequence of the Lucas numbers, i.e., we extend our previous work [5] and

determine all the solutions of the Diophantine equation L n +L m = a (1) in nonnegative integers n m and a. Similar problems have recently been investigated. For example, repdigits which are sums of at most three Fibonacci numbers were found by Luca [1]; Fibonacci numbers which are sums of two repdigits were obtained by Díaz and Luca [7], while factorials which are sums of at most three Fibonacci numbers were found by Luca and Siksek [11]. We prove the following result. Theorem. All solutions of the Diophantine equation (1) in nonnegative integers n m and a, are L 0 = 4, L 1 =, L 3 = 8, L +L 1 = 4, L 4 +L 1 = 8 and L 7 +L = 3. Let us give a brief overview of our strategy for proving Theorem. First, we rewrite equation (1) in suitable ways in order to obtain two different linear forms in logarithms which are both nonzero and small. Next, we use a lower bound on such nonzero linear forms in two logarithms due to Laurent, Mignotte, and Nesterenko as well as a general lower bound due to Matveev to find an absolute upper bound for n; hence, an absolute upper bound for m and a, which we then reduce using standard facts concerning continued fractions. In this paper, we follow the approach and the presentation described in [5]. Auxiliary results Before proceeding further, we recall that the Binet formula where L n = α n +β n holds for all n 0, α := 1+ 5 and β := 1 5 are the roots of the characteristic equation x x 1 = 0 of (L n ) n 0. This will be an important ingredient in what follows. In particular, the inequality α n 1 L n α n () holds for all n 0. In order to prove Theorem, we need a result of Laurent, Mignotte, and Nesterenko [10] about linear forms in two logarithms. But first, some notation. 3

Let η be an algebraic number of degree d with minimal polynomial d a 0 x d +a 1 x d 1 + +a d = a 0 (X η (i) ), where the a i s are relatively prime integers with a 0 > 0 and the η (i) s are conjugates of η. Then ( h(η) = 1 d loga 0 + log ( max{ η (i),1} )) d i=1 is called the logarithmic height of η. In particular, if η = p/q is a rational number with gcd(p,q) = 1 and q > 0, then h(η) = logmax{ p,q}. The following properties of the logarithmic height, which will be used in the next section without special reference, are also known: h(η ±γ) h(η)+h(γ)+log. h(ηγ ±1 ) h(η)+h(γ). h(η s ) = s h(η). With the above notation, Laurent, Mignotte, and Nesterenko [10, Corollary 1] proved the following theorem. Theorem 3. Let γ 1,γ be two non-zero algebraic numbers, and let logγ 1 and logγ be any determinations of their logarithms. Put D = [Q(γ 1,γ ) : Q]/[R(γ 1,γ ) : R], and i=1 Γ := b logγ b 1 logγ 1, where b 1 and b are positive integers. Further, let A 1,A be real numbers > 1 such that { loga i max h(γ i ), logγ i D, 1 }, i = 1,. D Then, assuming that γ 1 and γ are multiplicatively independent, we have where { log Γ > 30.9 D (max 4 logb, 1 D, 1 }) loga 1 loga, b = b 1 b +. DlogA DlogA 1 We shall also need the following general lower bound for linear forms in logarithms due to Matveev [13] (see also the paper of Bugeaud, Mignotte, and Siksek [3, Theorem 9.4]). 4

Theorem 4 (Matveev s theorem). Assume that γ 1,...,γ t are positive real algebraic numbers in a real algebraic number field K of degree D, b 1,...,b t are rational integers, and is not zero. Then where and Λ := γ b 1 1 γ bt t 1, Λ > exp ( 1.4 30 t+3 t 4.5 D (1+logD)(1+logB)A 1 A t ), B max{ b 1,..., b t }, A i max{dh(γ i ), logγ i,0.16}, for all i = 1,...,t. In 1998, Dujella and Pethő in [8, Lemma 5 (a)] gave a version of the reduction method based on the Baker-Davenport lemma [1]. To conclude this section of auxiliary results, we present the following lemma from [4], which is an immediate variation of the result due to Dujella and Pethő from [8], and will be one of the key tools used in this paper to reduce the upper bounds on the variables of the equation (1). Lemma 5. Let M be a positive integer, let p/q be a convergent of the continued fraction of the irrational γ such that q > 6M, and let A,B,µ be some real numbers with A > 0 and B > 1. Let ǫ := µq M γq, where denotes the distance from the nearest integer. If ǫ > 0, then there is no solution to the inequality in positive integers u,v and w with 0 < uγ v +µ < AB w, u M and w log(aq/ǫ) logb. 3 The Proof of Theorem Assume throughout that equation (1) holds. First of all, observe that if n = m, then the original equation (1) becomes L n = a 1. But the only solutions of this latter equation are (n,a) {(0,),(1,1),(3,3)} and this fact has already been mentioned in the Introduction. So, from now on, we assume that n > m. If n 00, then a brute force search with Mathematica in the range 0 m < n 00 gives the solutions (n,m,a) {(,1,),(4,1,3),(7,,5)}. This took a few seconds. Thus, for the rest of the paper we assume that n > 00. Let us now get a relation between n and a. Combining (1) with the right inequality of (), one gets that a α n +α m < n+1 + m+1 = n+1 (1+ m n ) n+1 (1+ 1 ) < n+, 5

which leads to a n+1. This estimate is essential for our purpose. On the other hand, we rewrite equation (1) as α n a = β n L m. We now take absolute values in the above relation obtaining α n a β n +L m < 1 +αm. Dividing both sides of the above expression by α n and taking into account that n > m, we get 1 a α n < 3 αn m. (3) In order to apply Theorem 3, we take γ 1 := α, γ :=, b 1 := n and b := a. So, Γ := b logγ b 1 logγ 1, and therefore estimation (3) can be rewritten as 1 e Γ < 3 α n m. (4) The algebraic number field containing γ 1,γ is Q( 5), so we can take D :=. By using (1) and the Binet formula for the Lucas sequence, we have α n = L n β n < L n +1 L n +L m = a. Consequently, 1 < a α n and so Γ > 0. This, together with (4), gives 0 < Γ < 3 αn m, (5) where we have also used the fact that x e x 1 for all x R. Hence, logγ < log3 (n m)logα. (6) Note further that h(γ 1 ) = (logα)/ = 0.406 and h(γ ) = log = 0.6931 ; thus, we can choose loga 1 := 0.5 and loga := 0.7. Finally, by recalling that a n+1, we get b = n +a < 1.7149n+1 < n. 1.4 Since α and are multiplicatively independent, we have, by Theorem 3, that logγ 30.9 4 (max{log(n),1/,1/}) 0.5 0.7 > 174 (max{log(n),1/,1/}). (7) 6

We now combine (6) and (7) to obtain (n m)logα < 180 (max{log(n),1/}). (8) Let us now get a second linear form in logarithms. To this end, we now rewrite (1) as follows: α n (1+α m n ) a = β n β m. Taking absolute values in the above relation and using the fact that β = (1 5)/, we get α n (1+α m n ) a = β n + β m < for all n > 00 and m 0. Dividing both sides of the above inequality by the first term of the left-hand side, we obtain 1 a α n (1+α m n ) 1 < αn. (9) We are now ready to apply Matveev s result Theorem 4. To do this, we take the parameters t := 3 and γ 1 :=, γ := α, γ 3 := 1+α m n. We take b 1 := a, b := n and b 3 := 1. As before, K := Q( 5) contains γ 1,γ,γ 3 and has D := [K : Q] =. To see why the left-hand side of (9) is not zero, note that otherwise, we would get the relation a = α n +α m. (10) Conjugating the above relation in Q( 5), we get Combining (10) and (11), we obtain a = β n +β m. (11) α n < α n +α m = β n +β m β n + β m <, which is impossible for n > 00. Hence, indeed the left-hand side of inequality (9) is nonzero. In thisapplication ofmatveev s theoremwe takea 1 := 1.4 anda := 0.5. Sincea n+1 it follows that we can take B := n+1. Let us now estimate h(γ 3 ). We begin by observing that γ 3 = 1+α m n < and γ 1 3 = 1 < 1, 1+αm n so that logγ 3 < 1. Next, notice that ( ) ( ) logα logα h(γ 3 ) m n +log = log+(n m). 7

Hence, we can take A 3 := +(n m)logα > max{h(γ 3 ), logγ 3,0.16}. Now Matveev s theorem implies that a lower bound on the left-hand side of (9) is exp( C (1+log(n+1)) 1.4 0.5 (+(n m)logα) where C := 1.4 30 6 3 4.5 (1+log) < 9.7 10 11. So, inequality (9) yields nlogα log < 1.36 10 1 logn (+(n m)logα), (1) where we used the inequality 1+log(n+1) < logn, which holds because n > 00. Using now (8) in the right-most term of the above inequality (1) and performing the respective calculations, we arrive at n < 6 10 14 logn (max{log(n),1/}). (13) If max{log(n),1/} = 1/, it then follows from (13) that n < 8 10 16 logn giving n < 3.5 10 18. If on the other hand we have that max{log(n),1/} = log(n), then, from (13), we get n < 6 10 14 lognlog (n) and so n < 5.9 10 19. In any case, we have that n < 5.9 10 19 always holds. We summarize what we have proved so far in the following lemma. Lemma 6. If (n,m,a) is a solution in positive integers of equation (1) with n > m and n > 00, then inequalities a n+1 < 6 10 19 hold. 4 Reducing the bound on n After finding an upper bound on n the next step is to reduce it. To do this, we first use some properties of continued fractions to obtain a suitable upper bound on n m, and secondly we use Lemma 5 to conclude that n must be relatively small. Let us see. Turning back to inequality (5), we obtain Dividing across by logα, we get 0 < alog nlogα < 3 α n m. 0 < aγ n < 7 log αn m, where γ := logα. (14) 8

Let [a 0,a 1,a,a 3,a 4,...] = [1,,3,1,,...] be the continued fraction expansion of γ, and let denote p k /q k its kth convergent. Recall also that a < 6 10 19 by Lemma 6. A quick inspection using Mathematica reveals that 54475119544877440894 = q 44 < 6 10 19 < q 45 = 671957765603468483. Furthermore, a M := max{a i i = 0,1,...,45} = a 17 = 134. So, from the known properties of continued fractions, we obtain that aγ n > Comparing estimates (14) and (15), we get right away that 1 (a M +)a. (15) α n m < 7 136 a < 6 10, leading to n m 110. Let us now work a little bit on (9) in order to find an improved upper bound on n. Put z := alog nlogα logϕ(n m), (16) where ϕ is the function given by the formula ϕ(t) := 1+α t. Therefore, (9) implies that 1 e z < αn. (17) Note that z 0; thus, we distinguish the following cases. If z > 0, then, from (17), we obtain 0 < z e z 1 < α n. Replacingz intheaboveinequalitybyitsformula(16)anddividingbothsidesoftheresulting inequality by logα, we get ( ) log 0 < a n logϕ(n m) < 5 α n. (18) logα logα We now put γ := log, µ := logϕ(n m), A := 5 and B := α. logα logα Clearlyγ isanirrationalnumber. WealsoputM := 6 10 19,whichisanupperboundonaby Lemma6. WethereforeapplyLemma5toinequality (18)forallchoicesn m {1,...,110} except when n m = 1,3 and get that n < log(aq/ǫ) logb, 9

where q > 6M is a denominator of a convergent of the continued fraction of γ such that ǫ = µq M γq > 0. Indeed, with the help of Mathematica we find that if (n,m,a) is a possible solution of the equation (1) with z > 0 and n m 1,3, then n 130. This is false because our assumption that n > 00. Suppose now that z < 0. First, note that /α n < 1/ since n > 00. Then, from (17), we have that 1 e z < 1/ and therefore e z <. Since z < 0, we have 0 < z e z 1 = e z e z 1 < 4 α n. Then, by the same arguments used for proving (18), we obtain ( ) logα 0 < n a+ logϕ(n m) < 6 α n. (19) log log Here, we also take M := 6 10 19, which is an upper bound on n by Lemma 6, and we apply Lemma 5 to inequality (19) for each n m {1,...,110} except for n m = 1,3. In this case, with the help of Mathematica, we find that if (n,m,a) is a possible solution of the equation (1) with z < 0 and n m 1,3, then n 10, which is false. Finally, we deal with the cases when n m = 1 and 3. We cannot study these cases as before because when applying Lemma 5 to the expressions (18) or (19) (according to whether z is positive or negative, respectively), the corresponding parameter µ appearing in Lemma 5 is either logϕ(t) logα = { 1, if t = 1; 1 log, if t = 3. or logα log ϕ(t) log = { logα, if t = 1; log, if t = 3. 1 logα log But, in any case, one can see that the corresponding value of ǫ from Lemma 5 is always negative and therefore the reduction method is not useful for reducing the bound on n in these instances. For this reason we need to treat these cases differently. All we want to do here is solve the equations L m+1 +L m = a and L m+3 +L m = a (0) in positive integers m and a with m + 1 > 00 and m + 3 > 00, respectively. But, by definition L m+1 +L m = L m+. Moreover, L m+3 +L m = L m+, which is easily checked. We see from the above discussion that equations (0) are transformed into the simpler equations L m+ = a and L m+ = a 1 (1) to be resolved in positive integers m and a with m > 199 and m > 197, respectively. But, we quickly see that the above equations (1) have no solutions for m > 1 as mentioned earlier. This completes the analysis of the cases when n m = 1,3 and therefore the proof of Theorem. 10

5 Acknowledgments We thank the referee for suggestions which improved the quality of this paper. J. J. B. was partially supported by University of Cauca and F. L. was supported in part by Project PAPIIT IN10451, UNAM, Mexico. References [1] A. Baker and H. Davenport, The equations 3x = y and 8x 7 = z, Quart. J. Math. Oxford Ser. 0 (1969), 19 137. [] Yu. Bilu, G. Hanrot, and P. Voutier, Existence of primitive divisors of Lucas and Lehmer numbers (with an appendix by M. Mignotte), J. Reine Angew. Math. 539 (001), 75 1. [3] Y. Bugeaud, M. Mignotte, and S. Siksek, Classical and modular approaches to exponential Diophantine equations. I. Fibonacci and Lucas perfect powers, Ann. of Math. 163 (006), 969 1018. [4] J. J. Bravo and F. Luca, On a conjecture about repdigits in k generalized Fibonacci sequences, Publ. Math. Debrecen 8 (013), 63 639. [5] J. J. Bravo and F. Luca, On the Diophantine equation F n +F m = a, preprint, 014. [6] R. D. Carmichael, On the numerical factors of the arithmetic forms α n ±β n, Ann. Math. 15 (1913), 30 70. [7] S. Díaz Alvarado and F. Luca, Fibonacci numbers which are sums of two repdigits, Proceedings of the XIVth International Conference on Fibonacci numbers and their applications, 011, pp. 97 111. [8] A. Dujella and A. Pethő, A generalization of a theorem of Baker and Davenport, Quart. J. Math. Oxford Ser. 49 (1998), no. 195, 91 306. [9] T. Koshy, Fibonacci and Lucas numbers with Applications, Wiley-Interscience Publications, 001. [10] M. Laurent, M. Mignotte, and Y. Nesterenko, Formes linéaires en deux logarithmes et déterminants d interpolation, (French) (Linear forms in two logarithms and interpolation determinants), J. Number Theory 55 (1995), 85 31. [11] F. Luca and S. Siksek, Factorials expressible as sums of two and three Fibonacci numbers, Proc. Edinb. Math. Soc. 53 (010), 747 763. 11

[1] F. Luca, Repdigits as sums of three Fibonacci numbers, Math. Commun. 17 (01), 1 11. [13] E. M. Matveev, An explicit lower bound for a homogeneous rational linear form in the logarithms of algebraic numbers, II, Izv. Ross. Akad. Nauk Ser. Mat. 64 (000), 15 180. Translation in Izv. Math. 64 (000), 117 169. 010 Mathematics Subject Classification: Primary 11B39; Secondary 11J86. Keywords: Fibonacci number, Lucas number, linear forms in logarithms, continued fraction, reduction method. (Concerned with sequences A00003 and A000045.) Received March 17 014; revised versions received July 014; July 30 014. Published in Journal of Integer Sequences, July 30 014. Return to Journal of Integer Sequences home page. 1